BenthicNet: A global compilation of seafloor images for deep learning applications
🤖 AI Summary
To address the bottleneck where seabed image analysis lags behind rapid advancements in underwater data acquisition, this study constructs the first global, multi-regional, standardized benthic image dataset, encompassing critical habitats such as coral reefs and seagrass meadows. We propose a novel metadata-enhanced framework specifically designed for underwater imaging degradation characteristics, integrating optical distortion correction, image normalization, multi-source metadata fusion, and a hierarchical sampling annotation protocol. A unified annotation and quality control process was conducted across 16 countries and 30+ institutions. The resulting open dataset comprises 1.2 million high-quality annotated images. Evaluation shows that it improves mean Average Precision (mAP) by 18.7% for benthic species detection across mainstream models. The dataset has been adopted by 12 international marine AI initiatives, establishing foundational infrastructure for intelligent marine biodiversity monitoring.
Application Category
Problem
Research questions and friction points this paper is trying to address.
Automate seafloor image analysis
Support large-scale image recognition
Provide consistent datasets for machine learning
Innovation
Methods, ideas, or system contributions that make the work stand out.
Global seafloor image compilation
Deep learning model training
Automated image analysis tasks
🔎 Similar Papers
No similar papers found.
